Providers

Seepient Agent supports multiple LLM providers with a unified interface. Bring your own API keys from each provider and switch seamlessly — all other code stays the same. Seepient Agent does not provide hosted LLM inference.

ProviderType

type ProviderType = "openai" | "anthropic" | "glm" | "openai-compatible";

Provider	Value	Default model
OpenAI	"openai"	gpt-5.4
Anthropic	"anthropic"	claude-sonnet-4-6-20260320
GLM	"glm"	opus
OpenAI-compatible	"openai-compatible"	gpt-5.4 (configurable)

Available models

OpenAI

Model ID	Display Name
gpt-5.4	GPT-5.4
gpt-5.4-pro	GPT-5.4 Pro
gpt-5.4-mini	GPT-5.4 Mini
gpt-5.4-nano	GPT-5.4 Nano
gpt-5.3-instant	GPT-5.3 Instant
gpt-5.3-codex	GPT-5.3 Codex
o3	o3
o3-mini	o3 Mini

Anthropic

Model ID	Display Name
claude-sonnet-4-6-20260320	Claude Sonnet 4.6
claude-opus-4-6-20260320	Claude Opus 4.6
claude-haiku-4-5-20251001	Claude Haiku 4.5

GLM

Alias	Model ID	Display Name
haiku	glm-4.5-air	GLM-4.5 Air
sonnet	glm-4.7	GLM-4.7
opus	glm-5.1	GLM-5.1

TIP

GLM accepts both the alias ("haiku", "sonnet", "opus") and the full model ID. Aliases are automatically resolved.

Quick usage

Pass model as an option:

import { askSeepient } from "seepient";

const result = await askSeepient("Explain recursion", {
  model: "claude-sonnet-4-6-20260320",
});

Provider credentials

Credentials live in provider management, never in the environment.

Credentials are stored through provider management (setup wizard, TUI dock, seepient auth login), never read from the environment. Embedders can inject their own CredentialStore (see below).

---

Embedder Catalog & Provider API

For multi-tenant applications and custom embedders, instantiate createSeepient to manage accounts, assignments, and query model catalogs:

import { createSeepient } from "seepient";

// In-memory isolated instance (e.g. per-tenant or test runner)
const seepient = await createSeepient({
  overlayFile: ":memory:",
});

// List distinct upstream providers (e.g. ["anthropic", "google", "openai"])
const providers = await seepient.listProviders();

// Inspect full available models catalog
const catalog = await seepient.getCatalog();
console.log(`Found ${catalog.length} available models across providers:`, providers);

// Add a provider account programmatically
await seepient.addProvider({
  accountId: "team_anthropic",
  upstreamProvider: "anthropic",
  credential: { mode: "paste", keyValue: process.env.MY_COMPANY_ANTHROPIC_KEY! }, // embedder-owned env is fine — Seepient never reads it
});

---

Per-Account Discovery with createProviderManagerApi

For fine-grained control over accounts and remote model discovery:

import { createProviderManagerApi } from "seepient";

const manager = createProviderManagerApi(runtime);

// Retrieve current configuration state and accounts
const state = await manager.getState();

// Refresh discovered models from upstream API for an account
const refreshResult = await manager.refreshModels("team_anthropic");
if (refreshResult.ok) {
  console.log(`Discovered ${refreshResult.discoveredCount} models`);
}

---

Runtime provider switching with agents

Use agent.switchProvider() to change the provider account (or model) mid-conversation:

import { createSeepient } from "seepient";

const agent = await createSeepient({
  model: "gpt-5.4",
});

// First turn with the default provider account
const r1 = await agent.chat("What is the capital of France?");
console.log(r1.text);

// Switch to another configured provider account for the next turn
await agent.switchProvider("team_anthropic", "claude-opus-4-6-20260320");

const r2 = await agent.chat("Tell me more about its history");
console.log(r2.text);

// Switch models within the default account
await agent.switchProvider("opus");

const r3 = await agent.chat("Summarize in Chinese");
console.log(r3.text);

TIP

switchProvider(account, model) targets a provider account from your configuration; with a single argument it switches the model only. The conversation history is fully preserved across switches.

---

OpenAI-compatible provider

OpenAI-protocol endpoints are configured like any other provider: add an account with a base URL through provider management (or a providers record from the SDK), with a stored key or none for keyless servers.

Wire format — one wire today. Every OpenAI-protocol account (upstreamProvider: "openai" or "openai-compatible") dials the OpenAI Responses API for chat turns: requests go to POST {baseUrl}/v1/responses, and your endpoint must serve that path. Chat-completions-only servers — most Ollama and vLLM deployments — are not usable for chat through this path today, whatever model names you declare. Model declarations affect assignment resolution (which names are allowed) but never the wire; the compat field is accepted by the schema but does not change wire selection. (Model discovery dials GET {baseUrl}/v1/models.) Chat-completions support is a known follow-up.

Related APIs

• Provider Management API -- Complete programmatic accounts, assignments, and runtime management
• createSeepient() -- Stateful agent with provider switching
• askSeepient() -- Stateless one-shot execution (streaming via stream: true)
• Custom Tools -- Register custom tools and trust boundaries
• Types Reference -- Full TypeScript type reference